On October 14, 2026, Labelbox announced more than $100 million in Series E financing, combining equity and a credit facility, at a $1.25 billion pre-money valuation. The round was led by Infinitum Partners, with Wipro Ventures, a16z, B Capital, Tau Ventures and Snowpoint Ventures participating.
In the announcement, CEO Manu Sharma described the company in three chapters. It began with tools for turning expert judgment into training data. After ChatGPT, it moved into reinforcement learning environments: through its Horizon product, frontier labs train models in simulated “worlds” built to mirror real work, with code, dashboards, spreadsheets and documents filled with realistic business data, where each step is recorded as a trajectory and graded by verifiers and experts. Labelbox says it now works with 90 percent of the leading frontier AI labs and hyperscalers and has more than 400 customers.
The third chapter is agents. Labelbox says more than 50 agents now do real work inside the company, and it has opened that system to customers as Recursion Managed Agents.
Why it matters: the training-data business is moving from labeling examples to building practice environments in which models learn by doing tasks. Labelbox’s pitch shows where the post-training money in frontier labs is going in 2026, and a data-labeling company is now selling both the environments that train agents and the agents themselves.
What it does not show: the “$100M+” mixes equity with a credit facility and the split is not disclosed, and the valuation is pre-money. The 90 percent and 400-customer figures are company claims with no revenue disclosed.